Fix support for hardware accelerated embedding generation via ollama (#2008)

* config changes

* fixing default settings:

* updating docs

Author: shanbady

docs/how-to/embeddings.md

@@ -52,17 +52,17 @@ To get setup:
 
 ```
 QDRANT_ENCODER=vector_search.encoders.litellm.LiteLLMEncoder
-LITELLM_API_BASE=http://docker.for.mac.host.internal:11434
+LITELLM_API_BASE=http://docker.for.mac.host.internal:11434/v1/
 QDRANT_DENSE_MODEL=<ollama model name>
 ```
 
-_Note_ - "LITELLM_API_BASE=http://docker.for.mac.host.internal:11434" is Mac specific - if you are using another OS you will need to figure out what your host machine's docker address is.
+_Note_ - "LITELLM_API_BASE=http://docker.for.mac.host.internal:11434/v1/" is Mac specific - if you are using another OS you will need to figure out what your host machine's docker address is.
 
 Sample .env file configuration on Mac:
 
 ```
 QDRANT_ENCODER=vector_search.encoders.litellm.LiteLLMEncoder
-LITELLM_API_BASE=http://docker.for.mac.host.internal:11434
+LITELLM_API_BASE=http://docker.for.mac.host.internal:11434/v1/
 QDRANT_DENSE_MODEL=all-minilm
 ```
 

main/settings.py

@@ -816,7 +816,7 @@ def get_all_config_keys():
 LITELLM_TOKEN_ENCODING_NAME = get_string(
     name="LITELLM_TOKEN_ENCODING_NAME", default=None
 )
-LITELLM_CUSTOM_PROVIDER = get_string(name="LITELLM_CUSTOM_PROVIDER", default="ollama")
+LITELLM_CUSTOM_PROVIDER = get_string(name="LITELLM_CUSTOM_PROVIDER", default="openai")
 LITELLM_API_BASE = get_string(name="LITELLM_API_BASE", default=None)
 
 

vector_search/encoders/litellm.py

@@ -28,11 +28,12 @@ def embed_documents(self, documents):
         return [result["embedding"] for result in self.get_embedding(documents)["data"]]
 
     def get_embedding(self, texts):
-        if settings.LITELLM_CUSTOM_PROVIDER and settings.LITELLM_API_BASE:
-            return embedding(
-                model=self.model_name,
-                input=texts,
-                api_base=settings.LITELLM_API_BASE,
-                custom_llm_provider=settings.LITELLM_CUSTOM_PROVIDER,
-            ).to_dict()
-        return embedding(model=self.model_name, input=texts).to_dict()
+        config = {
+            "model": self.model_name,
+            "input": texts,
+        }
+        if settings.LITELLM_CUSTOM_PROVIDER:
+            config["custom_llm_provider"] = settings.LITELLM_CUSTOM_PROVIDER
+        if settings.LITELLM_API_BASE:
+            config["api_base"] = settings.LITELLM_API_BASE
+        return embedding(**config).to_dict()